The machines that run artificial intelligence don’t sleep, and they can’t afford to lose power for even a moment.
That reality is reshaping the American energy business faster than most investors anticipated. At a Fletcher Lecture Luncheon at Hardin-Simmons University in Abilene, Texas, Kelcy Warren, Executive Chairman and co-founder of Energy Transfer, explained how AI data centers have become one of the most demanding customers the natural gas industry has ever seen, and why the infrastructure Energy Transfer spent three decades building turned out to be exactly what they need.
“These data centers are saying, ‘We can’t be interrupted,’” Warren told the audience. “‘We would lose billions of dollars of chips if we get interrupted.’ And air conditioning is what it amounts to.”
The comment captures a shift energy analysts have been tracking with growing urgency. The International Energy Agency projects global electricity consumption from data centers will roughly double: from 485 terawatt hours in 2025 to approximately 950 terawatt hours by 2030. Natural gas is the largest single source of additional supply expected to meet that load, set to add more than 130 terawatt hours of annual U.S. generation through the end of the decade.
Why Natural Gas Can’t Be Replaced at the Base
AI infrastructure has forced a reckoning with what reliable power actually requires. Data centers house thousands of GPUs, the processors that train and run large language models. They generate intense, continuous heat, and cooling that hardware demands uninterrupted electricity. Renewable sources such as wind and solar, which generate power intermittently, can’t guarantee that continuity on their own.
Warren was direct about the tradeoff. “You can’t count on that,” he said of renewable-only power approaches. “You’ve got to have natural gas base load for power or you don’t have reliable power.”
The numbers support that view. Natural gas planned capacity in the U.S. power sector grew from 11.1 percent of new additions in 2024 to 18.1 percent in 2026. Over that same period, non-renewable capacity additions surged 71 percent while planned renewable growth slowed to just 2 percent, a direct response to the reliability demands of the AI buildout. Warren put the macro trajectory plainly: “We’re going to take our power generation and double it in the next ten years. And as a nation. We’re going to double it. And we’re going to be right in the middle of that.”
How Redundancy Won the Contract
Being in the middle of that growth requires more than pipe in the ground. It requires the ability to guarantee delivery from multiple directions simultaneously, even when something goes wrong.
At the lecture, Kelcy Warren described an Energy Transfer contract to supply natural gas to a data center under development near the Hardin-Simmons campus. The company is also pursuing similar projects in Louisiana and outside Phoenix. Energy Transfer won none of those contracts on price alone.
“Our system, we can flow from north to south, south to north, east to west, west to east,” Warren said. “We can come at you from all directions. Well, if you take off a single line, a single pipeline, if one backhoe hits it and shuts it down, they’re going to lose every microchip they’ve got.”
That multi-directional flow capability was the decisive factor. “They chose Energy Transfer because of our redundancy and because of who we are,” Warren said. “I’m just really proud of what we’ve built and to supply gas here.”
Energy Transfer’s network of nearly 125,000 miles of pipelines spans the continent and moves approximately one-third of the country’s natural gas and crude oil. The reach of that system creates the redundancy data center operators now require: infrastructure Energy Transfer never designed with AI data centers in mind, yet precisely what the moment demanded.
Kelcy Warren, profiled by D CEO Magazine as the architect of one of America’s defining pipeline networks, described his core business in terms of a treadmill: “I’m in the business of moving a product that is depleting. Every day you’ve got to make that up.” The AI buildout has changed that equation. Data centers won’t allow interruption, and the pipelines built to supply them belong, in large part, to Energy Transfer.
